Fig 1.
Left: Fitness trajectories of the highest-scoring controllers evolved in each evolutionary run. Right: Distribution of the post-evaluation fitness scores achieved by the highest-scoring controller of the 30 evolutionary runs conducted. The box-plot summarises results from 400 post-evaluation samples.
Fig 2.
The figure summarises results of the average from the 400 post-evaluation samples of the highest-scoring controller in terms of the metrics presented in Section Metrics.
Left: The order metric trajectory across the 3000 simulation steps. Centre: box-plot of the distribution of the number of groups metric. Right: The swarm cohesion trajectory across the 3000 simulation steps. The shaded areas indicate the standard deviation.
Fig 3.
Example of the behaviour displayed by the highest-performing controller, with 5, 8, 11 and 16 robots, from left to right.
The lines represent the trajectory of the robots.
Fig 4.
Left: Fitness trajectories of the controllers in each of the 6000 generations. Right: Distribution of the post-evaluation fitness scores achieved by the highest-scoring controller.
Fig 5.
The metric scores achieved by the highest-scoring controllers of Global setup and the Local setup.
Left: the order metric. Centre: the number of groups metric. Right: the swarm cohesion metric. The shaded areas indicate the standard deviation.
Fig 6.
Example of the behaviour displayed by the highest-performing controller evolved in the Global setup, with 5, 8, 11 and 16 robots, from left to right.
Observing the beginning of trajectory, we can see the robots evolved the strategy of making a quick turn around their initial positions, in order to first form an aligned group before moving away.
Fig 7.
Example of the behaviour displayed in the scalability experiments, with 4, 8, 12, 20 and 40 robots, from left to right.
Fig 8.
Left: Fitness trajectories of the 30 controllers within 6000 generations. Right: Box-plot of the results of the 400 post-evaluation samples achieved by the highest-scoring controller of the No Alignment setup.
Fig 9.
The metric scores achieved by the highest-scoring controller of the No Alignment setup and the Global setup.
Left: The order metric. Centre: the number of groups metric. Right: the swarm cohesion metric. The shaded areas indicate the standard deviation.
Fig 10.
Example of the behaviour displayed by the highest-performing controller of the No Alignment setup, with 5, 8, 11 and 16 robots, from left to right.
The behaviour displayed is more cohesive than the one in Fig 6.